Brain-Computer Interface Project Overview

The BCI (brain-computer interface) Project is a multidisciplinary cooperation between the Department of Computer Engineering's NeuroTeam, the Department of Medical Psychology and Behavioral Neurobiology, and the Max-Planck Institute's Empirical Inference Department in T├╝bingen.

Persons who are completely paralysed by Amyotrophic Lateral Sclerosis lose the ability to communicate. The Medical Psychology Dept. supervises these patients and the BCI training required for them to regain communication skills with the help of their brain waves and the computer, while the Computer Engineering Dept. uses the skills and experience concentrated within the NeuroTeam to improve on state-of-the-art algorithms to speed up the computer's recognition of the paralysed patient's thoughts.

The BCI Project is ongoing, with the NeuroTeam having built up domain knowledge in the areas of feature extraction, feature selection, multi-class approaches, error potentials and cognition detection. We have gained knowledge in the use of recording methods as diverse as EEG, MEG and ECoG and the subsequent data processing specific to each of these methods. Together with its project partners, the NeuroTeam is in a unique position to direct its research effort toward the most afflicted patients.

A typical BCI setup is as follows: EEG electrodes are fixed to the patient's scalp. Potential differences due to electrical currents in the brain, originating from neural activity, are fed through an amplifier and into a computer. Algorithms are trained to recognise two conditions, such as imagined hand or foot movement, by repeated recordings (trials) of such imagination tasks done by the patient. This classification of two conditions allows the patient to choose letters or other elements on a computer screen, thus enabling communication between the patient and the outside world.

Feature Extraction and Selection

The project Machine Learning for automated Feature Extraction and Classification of EEG Signals for BCIs (AUMEX) was funded by the German Research Foundation DFG. It entailed the development of BCI systems which select the most important channels necessary for the classification algorithm to decode the patient's intent. Relevant publications are:

  • Schr├Âder, M., Lal, T. N., Hinterberger, T., Bogdan, M., Hill, N. J., Birbaumer, N., Rosenstiel, W. and Sch├Âlkopf, B.: Robust EEG Channel Selection Across Subjects for Brain Computer Interfaces. Journal on Applied Signal Processing, Special Issue: Trends in Brain Computer Interfaces. [pdf]
  • Schr├Âder, M., Bogdan, M., Rosenstiel, W., Hinterberger, T., and Birbaumer, N. (2003): Automated EEG Feature Selection for Brain Computer Interfaces. In: Proceedings of the 1st International IEEE EMBS Conference on Neural Engineering. pp. 626-629. March 20-22, 2003. [pdf]

Multi-class BCI

Our multi-class study with MEG recordings will provide evidence to support the idea of higher communication speed by the use of more than two imagination tasks. As this is our first study with up to seven classes, healthy subjects are participating to provide us with an insight into the increased cognitive load and the effect on classification accuracy when using multiple classes. An SVM classifier suited to the multi-class task combined with feature selection algorithms was constructed.

Error Potentials

Error potentials is currently a hot topic in BCI research. In the event of an incorrect letter being displayed to a person using a BCI for written communication, the user will be startled for a split-second. The "startled"-effect is characterised by negativity in the EEG signal and usually only brought to light when averaging over a large number of trials. In cooperation with Prof. Dr. A. K├╝bler of the W├╝rzburg University, the NeuroTeam is working on extracting this information buried deep within the signal in single trial patient data. This is a far harder task, but vital to increase the speed of communication in real-time.

Cognition Detection

Cognition detection tests based on EEG can be useful to determine time windows suitable for BCI training sessions in patients where vigilance is unclear. However, their interpretation is tricky as there can be no negative result. Even some healthy subjects do not exhibit the evoked potentials necessary to determine the outcome. We are developing Matlab tools to visualise and quantify cognition detection. Results from our ECoG patient study show a particularly high latency in the so-called P300 response to oddball tones. This evidence could point to slower overall cognition for the patient, which might be the reason for his failure to communicate via the BCI. In future patient studies, cognition detection could be used as a predictor of patient's BCI performance.


The research group of Prof. Wolfgang Rosenstiel will join forces with Prof. Niels Birbaumer, Institute of Medical Psychology and PD Alireza Gharabaghi (T├╝bingen University clinic for neurosurgery) to improve direct brain-computer communication within the BCCI project. The direction of information flow has traditionally been from the brain to the computer, and vice-versa only in the case of visual, auditory or tactile feedback. Information flow from the computer directly to the brain, via intelligent electrical stimulation, will be the main focus of research and is expected to improve the quality of communication. Stroke and ALS patients are targeted as beneficiaries of the new methods being developed.

Watch this space for more details soon.


Applications are important for the paralysed, as these allow the patients a more interesting form of conveying their intent than simply within the framework of a training session to gather data or deduce the current accuracy of the classifier. Applications open a wide range of possibilities to possibly increase the patients' quality of life.

Applications are spellers, games, interfaces to computer software or external devices. Computer software can be customised and optimised to the patients' requirements, such as the web browser Nessi which we are jointly developing with our project partners.

Relevant publications are:

  • Mellinger, J. and Hinterberger, T. and Bensch, M. and Schr├Âder M. and Birbaumer N. (2003): Surfing the Web with Electrical Brain Signals: The Brain Web Surfer (BWS) for the Completely Paralysed. Proceedings of 2nd World Congress of the International Society of Physical and Rehabilitation Medicine ISPRM Prague, Czech Republic, May 18-22, 2003. Edited by Haim Ring and Nachum Soroker. Download mellinger2003.pdf.gz.
  • E. Mugler, M. Bensch, S. Halder, W. Rosenstiel, M. Bogdan, N. Birbaumer, A. K├╝bler. Control of an Internet Browser Using the P300 Event Related Potential. International Journal of Bioelectromagnetism, Vol.10(1), p.56-63 [pdf]
  • M. Bensch, A. Karim, J. Mellinger, T. Hinterberger, M. Tangermann, M. Bogdan, W. Rosenstiel, N. Birbaumer (2008). Nessi: An EEG controlled web browser for severely paralyzed patients. Computational Intelligence and Neuroscience. [html]

Further Links

Neuroscience in T├╝bingen

Internet Browser Nessi